CCI + MFI Double Dip: Restored But Still On Watch

We tested a dual-oversold entry using CCI(20) below -80 and MFI(14) below 25, filtered to low-ADX periods and a 10-hour UTC trading window. The strategy was soft-killed in May 2026 after audit gates caught a -1.89 deflated Sharpe ratio and 100% fluke probability in walk-forward validation, then restored to observation status when regime analysis suggested the poor backtest numbers reflected a single hostile regime rather than a fundamentally broken logic.

The forward test ran 18 trades across BTC, ETH, XRP, and BNB on 5-minute bars. Average P&L per trade was 0.043% with a 55.6% win rate. Those are thin margins but positive, which is why the kill was reversed. The backtest told a different story: training and validation both averaged -0.232% per trade with 49.0% win rates across 1,273 out-of-sample trades. That contradiction is the core tension here.

Our hypothesis is that the strategy works in choppy, range-bound conditions (hence the ADX-below-30 filter) but got hammered during a trending regime that dominated the backtest period. The validation window may not have contained enough of the regime this strategy was designed for. We are rebuilding our regime detection to split results by market structure, which should clarify whether this is a niche tool or a statistical mirage.

Until that work completes, the gene remains in observation. It will not appear in the public survivor registry or selection pool, but it continues to paper trade so we accumulate more regime-tagged data. If you are considering a mean-reversion approach with CCI and MFI, treat this as a cautionary example: a strategy can pass forward testing and still fail rigorous cross-validation if the test window does not represent the conditions you are targeting.

Check the full survivor registry for strategies that passed all six gates without caveats, or head to StratProof Prove to backtest your own CCI and MFI thresholds against our multi-regime framework once the upgrade ships.